BadVPN – Rev 1

Subversion Repositories:
Rev:
.TH badvpn 7 "6 October 2010"
.SH NAME
BadVPN - peer-to-peer VPN system
.SH DESCRIPTION
.P
BadVPN is a peer-to-peer VPN system. It provides a Layer 2 (Ethernet) network between
the peers (VPN network nodes). The peers connect to a central server which acts as a chat
server for them to establish direct connections between each other (data connections).
These connections are used for transferring network data (Ethernet frames).
.SS "Features"
.P
.B "Data connections"
.P
Peers can transfer network data either over UDP or TCP. For both there are ways of
securing the data (see below).
.P
.B "IPv6 support"
.P
IPv6 can be used for both server connections and data connections, alongside with IPv4.
Additionally, both can be combined to allow gradual migration to IPv6.
.P
.B "Address selection"
.P
Because NATs and firewalls are widespread, it is harder for peer-to-peer services to operate.
In general, for two computers to be able to communicate, one computer must
.I bind
to one of its addresses, and the other computer must
.I connect
to the computer that binded (both for TCP and UDP). In a network with point-to-point
connectivity, the connecting computer can connect to the same address as the binding computer
bound to, so it is sufficient for the binding computer to send its address to the connecting
computer. However, NATs and firewalls break point-to-point connectivity. When a network is
behind a NAT, it is, by default, impossible for computers outside of that network to connect
to computers inside the network. This is because computers inside the network have no externally
visible IP address, and only communicate with the outside world through the external IP address
of the NAT router. It is however possible to manually configure the NAT router to
.I forward
a specific port number on its external IP address to a specific computer inside the network.
This makes it possible for a computer outside of the network to connect to a computer inside
a network, however, it must connect to the external address of the NAT router (rather than
the address the computer inside bound to, which is its internal address). So there needs
to be some way for the connecting peer to know what address to connect to.
.P
BadVPN solves this problem with so-called
.IR "address scopes" "."
The peer that binds must have a list of external addresses for each address it can bind to,
possibly ordered from best to worst. Each external address has its scope name. A scope name
represents part of a network from which an external address can be reached. On the other hand,
the peer that connects must have a list of scopes which it can reach. When a peer binds to an
address, it sends the other peer a list of external addresses along with scope names. That peer
than chooses the first external address whose scope it recognizes and attempts to connect to it
(if there is one).
.P
BadVPN also allows a peer to have multiple addresses for binding to. It is possible to specify
both an IPv4 and an IPv6 address to work in a multi-protocol environment.
.P
.B "Relaying"
.P
BadVPN can be configured to allow pairs of peers that cannot communicate directly (i.e. because of
NATs or firewalls) to relay network data through a third peer. Relaying is only attempted if
none of the two peers recognize any of the other peer's external addresses (or there are none).
For relaying to work, for each of the two peers (P1, other one P2) there must be at least one
third peer (R) that P1 it is allowed to relay through and can communicate directly with, and all
such peers R must be able to communicate directly with P2.
.P
.B "IGMP snooping"
.P
BadVPN nodes perform IGMP snooping in order to efficiently deliver multicast frames. For example,
this makes it possible to use BadVPN as a tunnel into an IPTV network of an Internet Service Provider
for you to watch TV from wherever you want (given sufficient link quality).
.P
.B "Code quality"
.P
BadVPN has great focus on code quality and reliability. BadVPN is written in the C programming
language. It is a single-threaded event-driven program. This allows for low resource usage and
fast response times. Even though C is a relatively low-level language, the programs are made of
small, highly cohesive and loosely coupled modules that are combined into a complete program on
a high level. Modules are accesed and communicate through small, simple and to-the-point interfaces.
It utilizes a flow-based design which greatly simplifies processing of data and input and output
of the programs.
.SS "Security features"
.P
BadVPN contains many security features, all of which are optional. The included security
features are described here.
.P
.B TLS for client-server connections
.P
It is possible for the peers to communicate with the chat server securely with TLS. It is
highly recommended that this feature is used if any security whatsoever is needed. Not
using it renders all other security features useless, since clients exchange keys
unencrypted via the server. When enabled, the chat server requires each client to identify
itself with a certificate.
.P
BadVPN uses Mozilla's NSS library for TLS support. This means that the required certificates
and keys must be available in a NSS database. The database and certificates can be
generated with the
.B certutil
command. See the examples section on how to generate and distribute the certificates.
.P
.B TLS for peer messaging
.P
If TLS is being used for client-server connections, it will also be used between each pair of
peers communicating via the server, on top of the TLS connections to the server. This secures
the messages from the server itself. It is important because the messages may include
encryption keys and other private data.
.P
.B TLS for TCP data connections
.P
If TCP is used for data connections between the peers, the data connections can be secured
with TLS. This requires using TLS for client-server connections. The clients need to trust
each others' certificates to be able to connect. Additionally, each client must identify to
its peers with the same certificates it used for connecting to the server.
.P
.B Encryption for UDP data connections
.P
If UDP is used for data connections, it is possible for each pair of peers to encrypt their
UDP packets with a symmetric block cipher. Note that the encryption keys are transmitted
through the server unencrypted, so for this to be useful, server connections must be secured
with TLS. The encryption aims to prevent third parties from seeing the real contents of
the network data being transfered.
.P
.B Hashes for UDP data connections
.P
If UDP is used for data connections, it is possible to include hashes in packets. Note that
hashes are only useful together with encryption. If enabled, the hash is calculated on the
packet with the hash field zeroed and then written to the hash field. Hashes are calculated
and included before encryption (if enabled). Combined with encryption, hashes aim to prevent
third parties from tampering with the packets and injecting them into the network.
.P
.B One-time passwords for UDP data connections
.P
If UDP is used for data connections, it is possible to include one-time passwords in packets.
Note that for this to be useful, server connections must be secured with TLS.
One-time passwords are generated from a seed value by encrypting zero data with a block cipher.
The seed contains the encryption key for the block cipher and the initialization vector.
Only a fixed number of passwords are used from a single seed. The peers exchange seeds through
the server. One-time passwords aim to prevent replay attacks.
.P
.B Control over peer communication
.P
It is possible to instruct the chat server to only allow certain peers to communicate. This
will break end-to-end connectivity in the virtual network. It is useful in certain cases
to improve security, for example when the VPN is used only to allow clients to securely connect
to a central service.
.SH "EXAMPLES"
.SS "Setting up certificates"
.P
If you want to use TLS for server connections (recommended), the server and all the peers will
need certificates. This section explains how to generate and distribute the certificates using
NSS command line tools.
.P
.B Setting up the Certificate Authority (CA)
.P
On the system that will host the CA, create a NSS database for the CA and generate a CA certificate
valid for 24 months:
.P
vpnca $ certutil -d sql:/home/vpnca/nssdb -N
.br
vpnca $ certutil -d sql:/home/vpnca/nssdb -S -n "vpnca" -s "CN=vpnca" -t "TC,," -x -2 -v 24
.br
> Is this a CA certificate [y/N]? y
.br
> Enter the path length constraint, enter to skip [<0 for unlimited path]: > -1
.br
> Is this a critical extension [y/N]? n
.P
Export the public CA certificate (this file is public):
.P
vpnca $ certutil -d sql:/home/vpnca/nssdb -L -n vpnca -a > ca.pem
.P
.B Setting up the server certificate
.P
On the CA system, generate a certificate for the server valid for 24 months, with TLS server usage context:
.P
vpnca $ certutil -d sql:/home/vpnca/nssdb -S -n "<insert_server_name>" -s "CN=<insert_server_name>" -c "vpnca" -t ",," -2 -6 -v 24
.br
> 0
.br
> -1
.br
> Is this a critical extension [y/N]? n
.br
> Is this a CA certificate [y/N]? n
.br
> Enter the path length constraint, enter to skip [<0 for unlimited path]: >
.br
> Is this a critical extension [y/N]? n
.P
Export the server certificate to a PKCS#12 file (this file must be kept secret):
.P
vpnca $ pk12util -d sql:/home/vpnca/nssdb -o server.p12 -n "<insert_server_name>"
.P
On the system that will run the server, create a NSS database and import the CA certificate
and the server cerificate:
.P
vpnserver $ certutil -d sql:/home/vpnserver/nssdb -N
.br
vpnserver $ certutil -d sql:/home/vpnserver/nssdb -A -t "CT,," -n "vpnca" -i /path/to/ca.pem
.br
vpnserver $ pk12util -d sql:/home/vpnserver/nssdb -i /path/to/server.p12
.P
.B Setting up peer certificates
.P
On the CA system, generate a certificate for the peer valid for 24 months, with TLS client and
TLS server usage contexts:
.P
vpnca $ certutil -d sql:/home/vpnca/nssdb -S -n "peer-<insert_name>" -s "CN=peer-<insert_name>" -c "vpnca" -t ",," -2 -6 -v 24
.br
> 0
.br
> 1
.br
> -1
.br
> Is this a critical extension [y/N]? n
.br
> Is this a CA certificate [y/N]? n
.br
> Enter the path length constraint, enter to skip [<0 for unlimited path]: >
.br
> Is this a critical extension [y/N]? n
.P
Export the peer certificate to a PKCS#12 file (this file must be kept secret):
.P
vpnca $ pk12util -d sql:/home/vpnca/nssdb -o peer-<insert_name>.p12 -n "peer-<insert_name>"
.P
On the system that will run the VPN client, create a NSS database and import the CA certificate
and the peer cerificate:
.P
vpnclient $ certutil -d sql:/home/vpnclient/nssdb -N
.br
vpnclient $ certutil -d sql:/home/vpnclient/nssdb -A -t "CT,," -n "vpnca" -i /path/to/ca.pem
.br
vpnclient $ pk12util -d sql:/home/vpnclient/nssdb -i /path/to/peer-<insert_name>.p12
.SS "Setting up TAP devices"
.P
You need to create and configure TAP devices on all computers that will participate in the virtual network
(i.e. run the client program). See
.BR badvpn-client (8),
section `TAP DEVICE CONFIGURATION` for details.
.SS "Example: Local IPv4 network, UDP transport, zero security"
.P
.B Starting the server:
.P
badvpn-server --listen-addr 0.0.0.0:7000
.P
.B Starting the peers:
.P
badvpn-client
.RS
--server-addr <insert_server_local_address>:7000
.br
--transport-mode udp --encryption-mode none --hash-mode none
.br
--scope local1
.br
--bind-addr 0.0.0.0:8000 --num-ports 30 --ext-addr {server_reported}:8000 local1
.br
--tapdev tap0
.RE
.SS "Example: Adding TLS and UDP security"
.P
.B Starting the server (other options as above):
.P
badvpn-server ...
.RS
--ssl --nssdb sql:/home/vpnserver/nssdb --server-cert-name "<insert_server_name>"
.RE
.P
.B Starting the peers (other options as above):
.P
badvpn-client ...
.RS
--ssl --nssdb sql:/home/vpnclient/nssdb --client-cert-name "peer-<insert_name>"
.br
--encryption-mode blowfish --hash-mode md5 --otp blowfish 3000 2000
.RE
.SS "Example: Multiple local networks behind NATs, all connected to the Internet"
.P
For each peer in the existing local network, configure the NAT router to forward its
range of ports to it (assuming their port ranges do not overlap). The clients also need
to know the external IP address of the NAT router. If you don't have a static one,
you'll need to discover it before starting the clients. Also forward the server port to
the server.
.P
.B Starting the peers in the local network (other options as above):
.P
badvpn-client
.RS
.RB "..."
.br
--scope internet
.br
.RB "..."
.br
--ext-addr <insert_NAT_routers_external_IP>:<insert_start_of_forwarded_port_range> internet
.br
.RB "..."
.RE
.P
The --ext-addr option applies to the previously specified --bind-addr option, and must come after
the first --ext-addr option which specifies a local address.
.P
Now perform a similar setup in some other local network behind a NAT. However:
.br
- Don't set up a new server, instead make the peers connect to the existing server in the first
local network.
.br
- You can't use {server_reported} for the local address --ext-addr options, because the server
would report the NAT router's external address rather than the peer's internal address. Instead
each peer has to know its internal IP address.
.br
- Use a different scope name for it, e.g. "local2" instead of "local1".
.P
If setup correctly, all peers will be able to communicate: those in the same local network will
communicate directly through local addresses, and those in different local networks will
communicate through the Internet.
.SH "PROTOCOL"
The protocols used in BadVPN are described in the source code in the protocol/ directory.
.SH "SEE ALSO"
.BR badvpn-server (8),
.BR badvpn-client (8)
.SH AUTHORS
Ambroz Bizjak <ambrop7@gmail.com>